replace free hugepage folios after migration

Commit Message

Ge Yang Dec. 18, 2024, 6:33 a.m. UTC

From: yangge <[e-mail]>

My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
have configured each NUMA node with 16GB of CMA and 16GB of in-use
hugetlb pages. The allocation of contiguous memory via the
cma_alloc() function can fail probabilistically.

The cma_alloc() function may fail if it sees an in-use hugetlb page
within the allocation range, even if that page has already been
migrated. When in-use hugetlb pages are migrated, they may simply
be released back into the free hugepage pool instead of being
returned to the buddy system. This can cause the
test_pages_isolated() function check to fail, ultimately leading
to the failure of the cma_alloc() function:
cma_alloc()
    __alloc_contig_migrate_range() // migrate in-use hugepage
    test_pages_isolated()
        __test_page_isolated_in_pageblock()
             PageBuddy(page) // check if the page is in buddy

To address this issue, we will add a function named
replace_free_hugepage_folios(). This function will replace the
hugepage in the free hugepage pool with a new one and release the
old one to the buddy system. After the migration of in-use hugetlb
pages is completed, we will invoke the replace_free_hugepage_folios()
function to ensure that these hugepages are properly released to
the buddy system. Following this step, when the test_pages_isolated()
function is executed for inspection, it will successfully pass.

Signed-off-by: yangge <[e-mail]>
---
 include/linux/hugetlb.h |  6 ++++++
 mm/hugetlb.c            | 37 +++++++++++++++++++++++++++++++++++++
 mm/page_alloc.c         | 13 ++++++++++++-
 3 files changed, 55 insertions(+), 1 deletion(-)

Comments

David Hildenbrand Dec. 19, 2024, 4:40 p.m. UTC | #1

On 18.12.24 07:33, [e-mail] wrote:
> From: yangge <[e-mail]>

CCing Oscar, who worked on migrating these pages during memory offlining 
and alloc_contig_range().

> 
> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
> have configured each NUMA node with 16GB of CMA and 16GB of in-use
> hugetlb pages. The allocation of contiguous memory via the
> cma_alloc() function can fail probabilistically.
> 
> The cma_alloc() function may fail if it sees an in-use hugetlb page
> within the allocation range, even if that page has already been
> migrated. When in-use hugetlb pages are migrated, they may simply
> be released back into the free hugepage pool instead of being
> returned to the buddy system. This can cause the
> test_pages_isolated() function check to fail, ultimately leading
> to the failure of the cma_alloc() function:
> cma_alloc()
>      __alloc_contig_migrate_range() // migrate in-use hugepage
>      test_pages_isolated()
>          __test_page_isolated_in_pageblock()
>               PageBuddy(page) // check if the page is in buddy

I thought this would be working as expected, at least we tested it with 
alloc_contig_range / virtio-mem a while ago.

On the memory_offlining path, we migrate hugetlb folios, but also 
dissolve any remaining free folios even if it means that we will going 
below the requested number of hugetlb pages in our pool.

During alloc_contig_range(), we only migrate them, to then free them up 
after migration.

Under which circumstances doe sit apply that "they may simply be 
released back into the free hugepage pool instead of being returned to 
the buddy system"?

> 
> To address this issue, we will add a function named
> replace_free_hugepage_folios(). This function will replace the
> hugepage in the free hugepage pool with a new one and release the
> old one to the buddy system. After the migration of in-use hugetlb
> pages is completed, we will invoke the replace_free_hugepage_folios()
> function to ensure that these hugepages are properly released to
> the buddy system. Following this step, when the test_pages_isolated()
> function is executed for inspection, it will successfully pass.
> 
> Signed-off-by: yangge <[e-mail]>
> ---
>   include/linux/hugetlb.h |  6 ++++++
>   mm/hugetlb.c            | 37 +++++++++++++++++++++++++++++++++++++
>   mm/page_alloc.c         | 13 ++++++++++++-
>   3 files changed, 55 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index ae4fe86..7d36ac8 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -681,6 +681,7 @@ struct huge_bootmem_page {
>   };
>   
>   int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   				unsigned long addr, int avoid_reserve);
>   struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
> @@ -1059,6 +1060,11 @@ static inline int isolate_or_dissolve_huge_page(struct page *page,
>   	return -ENOMEM;
>   }
>   
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
> +{
> +	return 0;
> +}
> +
>   static inline struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   					   unsigned long addr,
>   					   int avoid_reserve)
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 8e1db80..a099c54 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -2975,6 +2975,43 @@ int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list)
>   	return ret;
>   }
>   
> +/*
> + *  replace_free_hugepage_folios - Replace free hugepage folios in a given pfn
> + *  range with new folios.
> + *  @stat_pfn: start pfn of the given pfn range
> + *  @end_pfn: end pfn of the given pfn range
> + *  Returns 0 on success, otherwise negated error.
> + */
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
> +{
> +	struct hstate *h;
> +	struct folio *folio;
> +	int ret = 0;
> +
> +	LIST_HEAD(isolate_list);
> +
> +	while (start_pfn < end_pfn) {
> +		folio = pfn_folio(start_pfn);
> +		if (folio_test_hugetlb(folio)) {
> +			h = folio_hstate(folio);
> +		} else {
> +			start_pfn++;
> +			continue;
> +		}
> +
> +		if (!folio_ref_count(folio)) {
> +			ret = alloc_and_dissolve_hugetlb_folio(h, folio, &isolate_list);
> +			if (ret)
> +				break;
> +
> +			putback_movable_pages(&isolate_list);
> +		}
> +		start_pfn++;
> +	}
> +
> +	return ret;
> +}
> +
>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   				    unsigned long addr, int avoid_reserve)
>   {
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index dde19db..1dcea28 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -6504,7 +6504,18 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>   	ret = __alloc_contig_migrate_range(&cc, start, end, migratetype);
>   	if (ret && ret != -EBUSY)
>   		goto done;
> -	ret = 0;
> +
> +	/*
> +	 * When in-use hugetlb pages are migrated, they may simply be
> +	 * released back into the free hugepage pool instead of being
> +	 * returned to the buddy system. After the migration of in-use
> +	 * huge pages is completed, we will invoke the
> +	 * replace_free_hugepage_folios() function to ensure that
> +	 * these hugepages are properly released to the buddy system.
> +	 */
> +	ret = replace_free_hugepage_folios(start, end);
> +	if (ret)
> +		goto done;
>   
>   	/*
>   	 * Pages from [start, end) are within a pageblock_nr_pages

SeongJae Park Dec. 19, 2024, 6:43 p.m. UTC | #2

Hello,

On Wed, 18 Dec 2024 14:33:08 +0800 [e-mail] wrote:

[...]
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index ae4fe86..7d36ac8 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -681,6 +681,7 @@ struct huge_bootmem_page {
>  };
>  
>  int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
>  struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>  				unsigned long addr, int avoid_reserve);
>  struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
> @@ -1059,6 +1060,11 @@ static inline int isolate_or_dissolve_huge_page(struct page *page,
>  	return -ENOMEM;
>  }
>  
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
> +{
> +	return 0;
> +}
> +

I think this should be static inline.  Otherwise, build fails when
CONFIG_HUGETLB_PAGE is unset.  Since this is already merged into mm-unstable
and the problem and fix seems straigthforward, I directly sent my fix:
https://lore.kernel.org/[e-mail]


Thanks,
SJ

[...]

Ge Yang Dec. 20, 2024, 8:56 a.m. UTC | #3

在 2024/12/20 0:40, David Hildenbrand 写道:
> On 18.12.24 07:33, [e-mail] wrote:
>> From: yangge <[e-mail]>
> 
> CCing Oscar, who worked on migrating these pages during memory offlining 
> and alloc_contig_range().
> 
>>
>> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
>> have configured each NUMA node with 16GB of CMA and 16GB of in-use
>> hugetlb pages. The allocation of contiguous memory via the
>> cma_alloc() function can fail probabilistically.
>>
>> The cma_alloc() function may fail if it sees an in-use hugetlb page
>> within the allocation range, even if that page has already been
>> migrated. When in-use hugetlb pages are migrated, they may simply
>> be released back into the free hugepage pool instead of being
>> returned to the buddy system. This can cause the
>> test_pages_isolated() function check to fail, ultimately leading
>> to the failure of the cma_alloc() function:
>> cma_alloc()
>>      __alloc_contig_migrate_range() // migrate in-use hugepage
>>      test_pages_isolated()
>>          __test_page_isolated_in_pageblock()
>>               PageBuddy(page) // check if the page is in buddy
> 
> I thought this would be working as expected, at least we tested it with 
> alloc_contig_range / virtio-mem a while ago.
> 
> On the memory_offlining path, we migrate hugetlb folios, but also 
> dissolve any remaining free folios even if it means that we will going 
> below the requested number of hugetlb pages in our pool.
> 
> During alloc_contig_range(), we only migrate them, to then free them up 
> after migration.
> 
> Under which circumstances doe sit apply that "they may simply be 
> released back into the free hugepage pool instead of being returned to 
> the buddy system"?
> 

After migration, in-use hugetlb pages are only released back to the 
hugetlb pool and are not returned to the buddy system.

The specific steps for reproduction are as follows:
1，Reserve hugetlb pages. Some of these hugetlb pages are allocated 
within the CMA area.
echo 10240 > /proc/sys/vm/nr_hugepages

2，To ensure that hugetlb pages are in an in-use state, we can use the 
following command.
qemu-system-x86_64 \
   -mem-prealloc \
   -mem-path /dev/hugepage/ \
   ...

3，At this point, using cma_alloc() to allocate contiguous memory may 
result in a probable failure.

>>
>> To address this issue, we will add a function named
>> replace_free_hugepage_folios(). This function will replace the
>> hugepage in the free hugepage pool with a new one and release the
>> old one to the buddy system. After the migration of in-use hugetlb
>> pages is completed, we will invoke the replace_free_hugepage_folios()
>> function to ensure that these hugepages are properly released to
>> the buddy system. Following this step, when the test_pages_isolated()
>> function is executed for inspection, it will successfully pass.
>>
>> Signed-off-by: yangge <[e-mail]>
>> ---
>>   include/linux/hugetlb.h |  6 ++++++
>>   mm/hugetlb.c            | 37 +++++++++++++++++++++++++++++++++++++
>>   mm/page_alloc.c         | 13 ++++++++++++-
>>   3 files changed, 55 insertions(+), 1 deletion(-)
>>
>> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
>> index ae4fe86..7d36ac8 100644
>> --- a/include/linux/hugetlb.h
>> +++ b/include/linux/hugetlb.h
>> @@ -681,6 +681,7 @@ struct huge_bootmem_page {
>>   };
>>   int isolate_or_dissolve_huge_page(struct page *page, struct 
>> list_head *list);
>> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned 
>> long end_pfn);
>>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>>                   unsigned long addr, int avoid_reserve);
>>   struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int 
>> preferred_nid,
>> @@ -1059,6 +1060,11 @@ static inline int 
>> isolate_or_dissolve_huge_page(struct page *page,
>>       return -ENOMEM;
>>   }
>> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned 
>> long end_pfn)
>> +{
>> +    return 0;
>> +}
>> +
>>   static inline struct folio *alloc_hugetlb_folio(struct 
>> vm_area_struct *vma,
>>                          unsigned long addr,
>>                          int avoid_reserve)
>> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
>> index 8e1db80..a099c54 100644
>> --- a/mm/hugetlb.c
>> +++ b/mm/hugetlb.c
>> @@ -2975,6 +2975,43 @@ int isolate_or_dissolve_huge_page(struct page 
>> *page, struct list_head *list)
>>       return ret;
>>   }
>> +/*
>> + *  replace_free_hugepage_folios - Replace free hugepage folios in a 
>> given pfn
>> + *  range with new folios.
>> + *  @stat_pfn: start pfn of the given pfn range
>> + *  @end_pfn: end pfn of the given pfn range
>> + *  Returns 0 on success, otherwise negated error.
>> + */
>> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned 
>> long end_pfn)
>> +{
>> +    struct hstate *h;
>> +    struct folio *folio;
>> +    int ret = 0;
>> +
>> +    LIST_HEAD(isolate_list);
>> +
>> +    while (start_pfn < end_pfn) {
>> +        folio = pfn_folio(start_pfn);
>> +        if (folio_test_hugetlb(folio)) {
>> +            h = folio_hstate(folio);
>> +        } else {
>> +            start_pfn++;
>> +            continue;
>> +        }
>> +
>> +        if (!folio_ref_count(folio)) {
>> +            ret = alloc_and_dissolve_hugetlb_folio(h, folio, 
>> &isolate_list);
>> +            if (ret)
>> +                break;
>> +
>> +            putback_movable_pages(&isolate_list);
>> +        }
>> +        start_pfn++;
>> +    }
>> +
>> +    return ret;
>> +}
>> +
>>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>>                       unsigned long addr, int avoid_reserve)
>>   {
>> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
>> index dde19db..1dcea28 100644
>> --- a/mm/page_alloc.c
>> +++ b/mm/page_alloc.c
>> @@ -6504,7 +6504,18 @@ int alloc_contig_range_noprof(unsigned long 
>> start, unsigned long end,
>>       ret = __alloc_contig_migrate_range(&cc, start, end, migratetype);
>>       if (ret && ret != -EBUSY)
>>           goto done;
>> -    ret = 0;
>> +
>> +    /*
>> +     * When in-use hugetlb pages are migrated, they may simply be
>> +     * released back into the free hugepage pool instead of being
>> +     * returned to the buddy system. After the migration of in-use
>> +     * huge pages is completed, we will invoke the
>> +     * replace_free_hugepage_folios() function to ensure that
>> +     * these hugepages are properly released to the buddy system.
>> +     */
>> +    ret = replace_free_hugepage_folios(start, end);
>> +    if (ret)
>> +        goto done;
>>       /*
>>        * Pages from [start, end) are within a pageblock_nr_pages
> 
>

Ge Yang Dec. 20, 2024, 9:03 a.m. UTC | #4

在 2024/12/20 2:43, SeongJae Park 写道:
> Hello,
> 
> On Wed, 18 Dec 2024 14:33:08 +0800 [e-mail] wrote:
> 
> [...]
>> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
>> index ae4fe86..7d36ac8 100644
>> --- a/include/linux/hugetlb.h
>> +++ b/include/linux/hugetlb.h
>> @@ -681,6 +681,7 @@ struct huge_bootmem_page {
>>   };
>>   
>>   int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
>> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
>>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>>   				unsigned long addr, int avoid_reserve);
>>   struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
>> @@ -1059,6 +1060,11 @@ static inline int isolate_or_dissolve_huge_page(struct page *page,
>>   	return -ENOMEM;
>>   }
>>   
>> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
>> +{
>> +	return 0;
>> +}
>> +
> 
> I think this should be static inline.  Otherwise, build fails when
> CONFIG_HUGETLB_PAGE is unset.  Since this is already merged into mm-unstable
> and the problem and fix seems straigthforward, I directly sent my fix:
> https://lore.kernel.org/[e-mail]
> 
> 
> Thanks,
> SJ
> 
> [...]

Thanks.

David Hildenbrand Dec. 20, 2024, 4:30 p.m. UTC | #5

On 20.12.24 09:56, Ge Yang wrote:
> 
> 
> 在 2024/12/20 0:40, David Hildenbrand 写道:
>> On 18.12.24 07:33, [e-mail] wrote:
>>> From: yangge <[e-mail]>
>>
>> CCing Oscar, who worked on migrating these pages during memory offlining
>> and alloc_contig_range().
>>
>>>
>>> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
>>> have configured each NUMA node with 16GB of CMA and 16GB of in-use
>>> hugetlb pages. The allocation of contiguous memory via the
>>> cma_alloc() function can fail probabilistically.
>>>
>>> The cma_alloc() function may fail if it sees an in-use hugetlb page
>>> within the allocation range, even if that page has already been
>>> migrated. When in-use hugetlb pages are migrated, they may simply
>>> be released back into the free hugepage pool instead of being
>>> returned to the buddy system. This can cause the
>>> test_pages_isolated() function check to fail, ultimately leading
>>> to the failure of the cma_alloc() function:
>>> cma_alloc()
>>>       __alloc_contig_migrate_range() // migrate in-use hugepage
>>>       test_pages_isolated()
>>>           __test_page_isolated_in_pageblock()
>>>                PageBuddy(page) // check if the page is in buddy
>>
>> I thought this would be working as expected, at least we tested it with
>> alloc_contig_range / virtio-mem a while ago.
>>
>> On the memory_offlining path, we migrate hugetlb folios, but also
>> dissolve any remaining free folios even if it means that we will going
>> below the requested number of hugetlb pages in our pool.
>>
>> During alloc_contig_range(), we only migrate them, to then free them up
>> after migration.
>>
>> Under which circumstances doe sit apply that "they may simply be
>> released back into the free hugepage pool instead of being returned to
>> the buddy system"?
>>
> 
> After migration, in-use hugetlb pages are only released back to the
> hugetlb pool and are not returned to the buddy system.

We had

commit ae37c7ff79f1f030e28ec76c46ee032f8fd07607
Author: Oscar Salvador <[e-mail]>
Date:   Tue May 4 18:35:29 2021 -0700

     mm: make alloc_contig_range handle in-use hugetlb pages
     
     alloc_contig_range() will fail if it finds a HugeTLB page within the
     range, without a chance to handle them.  Since HugeTLB pages can be
     migrated as any LRU or Movable page, it does not make sense to bail out
     without trying.  Enable the interface to recognize in-use HugeTLB pages so
     we can migrate them, and have much better chances to succeed the call.


And I am trying to figure out if it never worked correctly, or if
something changed that broke it.


In start_isolate_page_range()->isolate_migratepages_block(), we do the

	ret = isolate_or_dissolve_huge_page(page, &cc->migratepages);

to add these folios to the cc->migratepages list.

In __alloc_contig_migrate_range(), we migrate the pages using migrate_pages().


After that, the src hugetlb folios should still be isolated? But I'm getting
confused when these pages get un-silated and putback to hugetlb/freed.


> 
> The specific steps for reproduction are as follows:
> 1，Reserve hugetlb pages. Some of these hugetlb pages are allocated
> within the CMA area.
> echo 10240 > /proc/sys/vm/nr_hugepages
> 
> 2，To ensure that hugetlb pages are in an in-use state, we can use the
> following command.
> qemu-system-x86_64 \
>     -mem-prealloc \
>     -mem-path /dev/hugepage/ \
>     ...
> 
> 3，At this point, using cma_alloc() to allocate contiguous memory may
> result in a probable failure.
>

Will these free hugetlb folios become surplus pages? I would have assumed
they get freed immediately to the buddy, or does you config maybe allow for
surplus pages?

Ge Yang Dec. 21, 2024, 12:04 p.m. UTC | #6

在 2024/12/21 0:30, David Hildenbrand 写道:
> On 20.12.24 09:56, Ge Yang wrote:
>>
>>
>> 在 2024/12/20 0:40, David Hildenbrand 写道:
>>> On 18.12.24 07:33, [e-mail] wrote:
>>>> From: yangge <[e-mail]>
>>>
>>> CCing Oscar, who worked on migrating these pages during memory offlining
>>> and alloc_contig_range().
>>>
>>>>
>>>> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
>>>> have configured each NUMA node with 16GB of CMA and 16GB of in-use
>>>> hugetlb pages. The allocation of contiguous memory via the
>>>> cma_alloc() function can fail probabilistically.
>>>>
>>>> The cma_alloc() function may fail if it sees an in-use hugetlb page
>>>> within the allocation range, even if that page has already been
>>>> migrated. When in-use hugetlb pages are migrated, they may simply
>>>> be released back into the free hugepage pool instead of being
>>>> returned to the buddy system. This can cause the
>>>> test_pages_isolated() function check to fail, ultimately leading
>>>> to the failure of the cma_alloc() function:
>>>> cma_alloc()
>>>>       __alloc_contig_migrate_range() // migrate in-use hugepage
>>>>       test_pages_isolated()
>>>>           __test_page_isolated_in_pageblock()
>>>>                PageBuddy(page) // check if the page is in buddy
>>>
>>> I thought this would be working as expected, at least we tested it with
>>> alloc_contig_range / virtio-mem a while ago.
>>>
>>> On the memory_offlining path, we migrate hugetlb folios, but also
>>> dissolve any remaining free folios even if it means that we will going
>>> below the requested number of hugetlb pages in our pool.
>>>
>>> During alloc_contig_range(), we only migrate them, to then free them up
>>> after migration.
>>>
>>> Under which circumstances doe sit apply that "they may simply be
>>> released back into the free hugepage pool instead of being returned to
>>> the buddy system"?
>>>
>>
>> After migration, in-use hugetlb pages are only released back to the
>> hugetlb pool and are not returned to the buddy system.
> 
> We had
> 
> commit ae37c7ff79f1f030e28ec76c46ee032f8fd07607
> Author: Oscar Salvador <[e-mail]>
> Date:   Tue May 4 18:35:29 2021 -0700
> 
>      mm: make alloc_contig_range handle in-use hugetlb pages
>      alloc_contig_range() will fail if it finds a HugeTLB page within the
>      range, without a chance to handle them.  Since HugeTLB pages can be
>      migrated as any LRU or Movable page, it does not make sense to bail 
> out
>      without trying.  Enable the interface to recognize in-use HugeTLB 
> pages so
>      we can migrate them, and have much better chances to succeed the call.
> 
> 
> And I am trying to figure out if it never worked correctly, or if
> something changed that broke it.
> 
> 
> In start_isolate_page_range()->isolate_migratepages_block(), we do the
> 
>      ret = isolate_or_dissolve_huge_page(page, &cc->migratepages);
> 
> to add these folios to the cc->migratepages list.
> 
> In __alloc_contig_migrate_range(), we migrate the pages using 
> migrate_pages().
> 
> 
> After that, the src hugetlb folios should still be isolated? 
Yes.

But I'm
> getting
> confused when these pages get un-silated and putback to hugetlb/freed.
> 
If the migration is successful, call folio_putback_active_hugetlb to 
release the src hugetlb folios back to the free hugetlb pool.

trace:
unmap_and_move_huge_page
     folio_putback_active_hugetlb
         folio_put
             free_huge_folio

alloc_contig_range_noprof
     __alloc_contig_migrate_range
     if (test_pages_isolated())  //to determine if hugetlb pages in buddy
         isolate_freepages_range //grab isolated pages from freelists.
     else
         undo_isolate_page_range //undo isolate

> 
>>
>> The specific steps for reproduction are as follows:
>> 1，Reserve hugetlb pages. Some of these hugetlb pages are allocated
>> within the CMA area.
>> echo 10240 > /proc/sys/vm/nr_hugepages
>>
>> 2，To ensure that hugetlb pages are in an in-use state, we can use the
>> following command.
>> qemu-system-x86_64 \
>>     -mem-prealloc \
>>     -mem-path /dev/hugepage/ \
>>     ...
>>
>> 3，At this point, using cma_alloc() to allocate contiguous memory may
>> result in a probable failure.
>>
> 
> Will these free hugetlb folios become surplus pages? I would have assumed
> they get freed immediately to the buddy, or does you config maybe allow for
> surplus pages?
> 
These freed hugetlb folios will not become surplus pages. I have not 
configured the system to allow for the existence of surplus pages.

David Hildenbrand Dec. 21, 2024, 2:32 p.m. UTC | #7

On 21.12.24 13:04, Ge Yang wrote:
> 
> 
> 在 2024/12/21 0:30, David Hildenbrand 写道:
>> On 20.12.24 09:56, Ge Yang wrote:
>>>
>>>
>>> 在 2024/12/20 0:40, David Hildenbrand 写道:
>>>> On 18.12.24 07:33, [e-mail] wrote:
>>>>> From: yangge <[e-mail]>
>>>>
>>>> CCing Oscar, who worked on migrating these pages during memory offlining
>>>> and alloc_contig_range().
>>>>
>>>>>
>>>>> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
>>>>> have configured each NUMA node with 16GB of CMA and 16GB of in-use
>>>>> hugetlb pages. The allocation of contiguous memory via the
>>>>> cma_alloc() function can fail probabilistically.
>>>>>
>>>>> The cma_alloc() function may fail if it sees an in-use hugetlb page
>>>>> within the allocation range, even if that page has already been
>>>>> migrated. When in-use hugetlb pages are migrated, they may simply
>>>>> be released back into the free hugepage pool instead of being
>>>>> returned to the buddy system. This can cause the
>>>>> test_pages_isolated() function check to fail, ultimately leading
>>>>> to the failure of the cma_alloc() function:
>>>>> cma_alloc()
>>>>>        __alloc_contig_migrate_range() // migrate in-use hugepage
>>>>>        test_pages_isolated()
>>>>>            __test_page_isolated_in_pageblock()
>>>>>                 PageBuddy(page) // check if the page is in buddy
>>>>
>>>> I thought this would be working as expected, at least we tested it with
>>>> alloc_contig_range / virtio-mem a while ago.
>>>>
>>>> On the memory_offlining path, we migrate hugetlb folios, but also
>>>> dissolve any remaining free folios even if it means that we will going
>>>> below the requested number of hugetlb pages in our pool.
>>>>
>>>> During alloc_contig_range(), we only migrate them, to then free them up
>>>> after migration.
>>>>
>>>> Under which circumstances doe sit apply that "they may simply be
>>>> released back into the free hugepage pool instead of being returned to
>>>> the buddy system"?
>>>>
>>>
>>> After migration, in-use hugetlb pages are only released back to the
>>> hugetlb pool and are not returned to the buddy system.
>>
>> We had
>>
>> commit ae37c7ff79f1f030e28ec76c46ee032f8fd07607
>> Author: Oscar Salvador <[e-mail]>
>> Date:   Tue May 4 18:35:29 2021 -0700
>>
>>       mm: make alloc_contig_range handle in-use hugetlb pages
>>       alloc_contig_range() will fail if it finds a HugeTLB page within the
>>       range, without a chance to handle them.  Since HugeTLB pages can be
>>       migrated as any LRU or Movable page, it does not make sense to bail
>> out
>>       without trying.  Enable the interface to recognize in-use HugeTLB
>> pages so
>>       we can migrate them, and have much better chances to succeed the call.
>>
>>
>> And I am trying to figure out if it never worked correctly, or if
>> something changed that broke it.
>>
>>
>> In start_isolate_page_range()->isolate_migratepages_block(), we do the
>>
>>       ret = isolate_or_dissolve_huge_page(page, &cc->migratepages);
>>
>> to add these folios to the cc->migratepages list.
>>
>> In __alloc_contig_migrate_range(), we migrate the pages using
>> migrate_pages().
>>
>>
>> After that, the src hugetlb folios should still be isolated?
> Yes.
> 
> But I'm
>> getting
>> confused when these pages get un-silated and putback to hugetlb/freed.
>>
> If the migration is successful, call folio_putback_active_hugetlb to
> release the src hugetlb folios back to the free hugetlb pool.
> 
> trace:
> unmap_and_move_huge_page
>       folio_putback_active_hugetlb
>           folio_put
>               free_huge_folio
> 
> alloc_contig_range_noprof
>       __alloc_contig_migrate_range
>       if (test_pages_isolated())  //to determine if hugetlb pages in buddy
>           isolate_freepages_range //grab isolated pages from freelists.
>       else
>           undo_isolate_page_range //undo isolate

Ah, now I remember, thanks.

So when we free an ordinary page, we put it onto the buddy isolate list, 
from where we can grab it later and nobody can allocate it in the meantime.

In case of hugetlb, we simply free it back to hugetlb, from where it can 
likely even get allocated immediately again.

I think that can actually happen in your proposal: the now-free page 
will get reallocated, for example for migrating the next folio. Or some 
concurrent system activity can simply allocate the now-free folio. Or am 
I missing something that prevents these now-free hugetlb folios from 
getting re-allocated after migration succeeded?


Conceptually, I think we would want migration code in the case of 
alloc_contig_range() to allocate a new folio from the buddy, and to free 
the old one back to the buddy immediately, without ever allowing 
re-allocation of it.

What needs to be handled is detecting that

(a) we want to allocate a fresh hugetlb folio as migration target
(b) if migration succeeds, we have to free the hugetlb folio back to the 
buddy
(c) if migation fails, we have to free the allocated hugetlb foliio back 
to the buddy


We could provide a custom alloc_migration_target that we pass to 
migrate_page to allocate a fresh hugetlb folio to handle (a). Using the 
put_new_folio callback we could handle (c). (b) would need some thought.

Maybe we can also just mark the source folio as we isolate it, and 
enlighten migration+freeing code to handle it automatically?

Hoping to get some feedback from hugetlb maintainers.

David Hildenbrand Dec. 21, 2024, 2:35 p.m. UTC | #8

On 18.12.24 07:33, [e-mail] wrote:
> From: yangge <[e-mail]>
> 
> My machine has 4 NUMA nodes, each equipped with 32GB of memory. I
> have configured each NUMA node with 16GB of CMA and 16GB of in-use
> hugetlb pages. The allocation of contiguous memory via the
> cma_alloc() function can fail probabilistically.
> 
> The cma_alloc() function may fail if it sees an in-use hugetlb page
> within the allocation range, even if that page has already been
> migrated. When in-use hugetlb pages are migrated, they may simply
> be released back into the free hugepage pool instead of being
> returned to the buddy system. This can cause the
> test_pages_isolated() function check to fail, ultimately leading
> to the failure of the cma_alloc() function:
> cma_alloc()
>      __alloc_contig_migrate_range() // migrate in-use hugepage
>      test_pages_isolated()
>          __test_page_isolated_in_pageblock()
>               PageBuddy(page) // check if the page is in buddy
> 
> To address this issue, we will add a function named
> replace_free_hugepage_folios(). This function will replace the
> hugepage in the free hugepage pool with a new one and release the
> old one to the buddy system. After the migration of in-use hugetlb
> pages is completed, we will invoke the replace_free_hugepage_folios()
> function to ensure that these hugepages are properly released to
> the buddy system. Following this step, when the test_pages_isolated()
> function is executed for inspection, it will successfully pass.
> 
> Signed-off-by: yangge <[e-mail]>
> ---
>   include/linux/hugetlb.h |  6 ++++++
>   mm/hugetlb.c            | 37 +++++++++++++++++++++++++++++++++++++
>   mm/page_alloc.c         | 13 ++++++++++++-
>   3 files changed, 55 insertions(+), 1 deletion(-)
> 
> diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
> index ae4fe86..7d36ac8 100644
> --- a/include/linux/hugetlb.h
> +++ b/include/linux/hugetlb.h
> @@ -681,6 +681,7 @@ struct huge_bootmem_page {
>   };
>   
>   int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   				unsigned long addr, int avoid_reserve);
>   struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
> @@ -1059,6 +1060,11 @@ static inline int isolate_or_dissolve_huge_page(struct page *page,
>   	return -ENOMEM;
>   }
>   
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
> +{
> +	return 0;
> +}
> +
>   static inline struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   					   unsigned long addr,
>   					   int avoid_reserve)
> diff --git a/mm/hugetlb.c b/mm/hugetlb.c
> index 8e1db80..a099c54 100644
> --- a/mm/hugetlb.c
> +++ b/mm/hugetlb.c
> @@ -2975,6 +2975,43 @@ int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list)
>   	return ret;
>   }
>   
> +/*
> + *  replace_free_hugepage_folios - Replace free hugepage folios in a given pfn
> + *  range with new folios.
> + *  @stat_pfn: start pfn of the given pfn range
> + *  @end_pfn: end pfn of the given pfn range
> + *  Returns 0 on success, otherwise negated error.
> + */
> +int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
> +{
> +	struct hstate *h;
> +	struct folio *folio;
> +	int ret = 0;
> +
> +	LIST_HEAD(isolate_list);
> +
> +	while (start_pfn < end_pfn) {
> +		folio = pfn_folio(start_pfn);
> +		if (folio_test_hugetlb(folio)) {
> +			h = folio_hstate(folio);
> +		} else {
> +			start_pfn++;
> +			continue;
> +		}
> +
> +		if (!folio_ref_count(folio)) {
> +			ret = alloc_and_dissolve_hugetlb_folio(h, folio, &isolate_list);
> +			if (ret)
> +				break;
> +
> +			putback_movable_pages(&isolate_list);
> +		}
> +		start_pfn++;
> +	}
> +
> +	return ret;
> +}
> +
>   struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
>   				    unsigned long addr, int avoid_reserve)
>   {
> diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> index dde19db..1dcea28 100644
> --- a/mm/page_alloc.c
> +++ b/mm/page_alloc.c
> @@ -6504,7 +6504,18 @@ int alloc_contig_range_noprof(unsigned long start, unsigned long end,
>   	ret = __alloc_contig_migrate_range(&cc, start, end, migratetype);
>   	if (ret && ret != -EBUSY)
>   		goto done;
> -	ret = 0;
> +
> +	/*
> +	 * When in-use hugetlb pages are migrated, they may simply be
> +	 * released back into the free hugepage pool instead of being
> +	 * returned to the buddy system. After the migration of in-use
> +	 * huge pages is completed, we will invoke the
> +	 * replace_free_hugepage_folios() function to ensure that
> +	 * these hugepages are properly released to the buddy system.
> +	 */

As mentioned in my other mail, what I don't like about this is, IIUC, 
the pages can get reallocated anytime after we successfully migrated 
them, or is there  anything that prevents that?

Did you ever try allocating a larger range with a single 
alloc_contig_range() call, that possibly has to migrate multiple hugetlb 
folios in one go (and maybe just allocates one of the just-freed hugetlb 
folios as migration target)?

Patch

diff --git a/include/linux/hugetlb.h b/include/linux/hugetlb.h
index ae4fe86..7d36ac8 100644
--- a/include/linux/hugetlb.h
+++ b/include/linux/hugetlb.h
@@ -681,6 +681,7 @@  struct huge_bootmem_page {
 };
 
 int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list);
+int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn);
 struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 				unsigned long addr, int avoid_reserve);
 struct folio *alloc_hugetlb_folio_nodemask(struct hstate *h, int preferred_nid,
@@ -1059,6 +1060,11 @@  static inline int isolate_or_dissolve_huge_page(struct page *page,
 	return -ENOMEM;
 }
 
+int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
+{
+	return 0;
+}
+
 static inline struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 					   unsigned long addr,
 					   int avoid_reserve)
diff --git a/mm/hugetlb.c b/mm/hugetlb.c
index 8e1db80..a099c54 100644
--- a/mm/hugetlb.c
+++ b/mm/hugetlb.c
@@ -2975,6 +2975,43 @@  int isolate_or_dissolve_huge_page(struct page *page, struct list_head *list)
 	return ret;
 }
 
+/*
+ *  replace_free_hugepage_folios - Replace free hugepage folios in a given pfn
+ *  range with new folios.
+ *  @stat_pfn: start pfn of the given pfn range
+ *  @end_pfn: end pfn of the given pfn range
+ *  Returns 0 on success, otherwise negated error.
+ */
+int replace_free_hugepage_folios(unsigned long start_pfn, unsigned long end_pfn)
+{
+	struct hstate *h;
+	struct folio *folio;
+	int ret = 0;
+
+	LIST_HEAD(isolate_list);
+
+	while (start_pfn < end_pfn) {
+		folio = pfn_folio(start_pfn);
+		if (folio_test_hugetlb(folio)) {
+			h = folio_hstate(folio);
+		} else {
+			start_pfn++;
+			continue;
+		}
+
+		if (!folio_ref_count(folio)) {
+			ret = alloc_and_dissolve_hugetlb_folio(h, folio, &isolate_list);
+			if (ret)
+				break;
+
+			putback_movable_pages(&isolate_list);
+		}
+		start_pfn++;
+	}
+
+	return ret;
+}
+
 struct folio *alloc_hugetlb_folio(struct vm_area_struct *vma,
 				    unsigned long addr, int avoid_reserve)
 {
diff --git a/mm/page_alloc.c b/mm/page_alloc.c
index dde19db..1dcea28 100644
--- a/mm/page_alloc.c
+++ b/mm/page_alloc.c
@@ -6504,7 +6504,18 @@  int alloc_contig_range_noprof(unsigned long start, unsigned long end,
 	ret = __alloc_contig_migrate_range(&cc, start, end, migratetype);
 	if (ret && ret != -EBUSY)
 		goto done;
-	ret = 0;
+
+	/*
+	 * When in-use hugetlb pages are migrated, they may simply be
+	 * released back into the free hugepage pool instead of being
+	 * returned to the buddy system. After the migration of in-use
+	 * huge pages is completed, we will invoke the
+	 * replace_free_hugepage_folios() function to ensure that
+	 * these hugepages are properly released to the buddy system.
+	 */
+	ret = replace_free_hugepage_folios(start, end);
+	if (ret)
+		goto done;
 
 	/*
 	 * Pages from [start, end) are within a pageblock_nr_pages